Claude Fable 5 Release (2026): Anthropic's Most Powerful AI Model Explained

Claude Fable 5 is the most powerful AI model Anthropic has ever made available to the public. It launched on June 9, 2026, as the first “Mythos-class” model that anyone can use, not just a small group of vetted partners.

Mythos-class is a new tier that sits above Anthropic’s Opus family. The company said Fable 5 is state-of-the-art on nearly every tested benchmark. It shines in software engineering, complex knowledge work, vision tasks, and scientific research. The longer and harder the task, the bigger its lead over earlier Claude models.

But this is not a simple “here is a smarter chatbot” launch. Anthropic built an entire safety layer around the model because it is so capable. For most users, this is invisible. For a small share of queries, it matters a lot.

Last updated: June 9, 2026. Based on Anthropic’s official announcement and early access partner reports. For the full model landscape, see our latest AI models compared hub page.

Quick specs

The Mythos-class tier explained

Anthropic has used a naming system with tiers: Haiku (fast and cheap), Sonnet (balanced), and Opus (most capable). Mythos-class is a new tier above Opus.

The first Mythos-class model, Claude Mythos Preview, shipped in April 2026 through Project Glasswing. Only a small group of cybersecurity defenders and critical infrastructure providers got access. At the time, Anthropic said the models were too dangerous to release publicly because of their cybersecurity capabilities.

Fable 5 changes that. Anthropic spent months building safeguards that can detect and block high-risk queries reliably enough to release the model to everyone. The name “Fable” comes from the Latin fabula, meaning “that which is told.” Anthropic noted it is similar to the Greek mythos, reflecting that both models share the same underlying intelligence.

Claude Mythos 5 launched the same day. It is the exact same model, but with cybersecurity safeguards lifted. It is restricted to Project Glasswing partners for now.

What Fable 5 can actually do

Anthropic’s announcement focused on four areas where Fable 5 shows the biggest gains over previous models.

Software engineering

The standout result came from Stripe. In early testing, Fable 5 performed a codebase-wide migration across a 50-million-line Ruby codebase in a single day. Stripe said the same work would have taken a whole team over two months by hand.

On Cognition’s FrontierCode evaluation, which tests whether models can pass difficult coding tasks while meeting production codebase standards, Fable 5 scored highest among all frontier models at medium effort. It also handles coding tasks with fewer tokens than earlier Claude models, which means lower API costs for the same output.

Cursor put Fable 5 at the top of its CursorBench evaluation. GitHub’s chief product officer called it “a real step forward for the developers GitHub serves.”

Knowledge work and reasoning

On Hebbia’s Finance Benchmark for senior-level reasoning, Fable 5 earned the highest score of any model. It showed major gains in document-based reasoning, chart and table interpretation, and problem solving. IMC, a trading firm, said Fable 5 aced their trading analysis evaluations across the board.

One early access partner summarized the shift: “Fable 5’s reasoning is a clear step beyond Opus 4.8. It works at senior research scientist grade, picking directions, allocating resources, killing its incorrect beliefs, and producing novel first-principles outputs.”

Vision

Fable 5 is the new state-of-the-art model for vision tasks. It can pull precise numbers from detailed scientific figures. It can rebuild a web app’s entire source code from screenshots alone.

Anthropic tested it on Pokémon FireRed, a game that earlier Claude models could not play without extra tooling and game-state helpers. Fable 5 completed the game using only raw screenshots, with no maps, navigation aids, or extra information. That is a before-and-after picture of what vision in AI models can now do.

Memory and long-context handling

Fable 5 stays focused across millions of tokens and improves its outputs using its own notes. When Anthropic had the model play Slay the Spire, a deck-building game, giving it access to persistent file-based memory improved its performance three times more than the same upgrade did for Opus 4.8. Fable 5 also reached the game’s final act three times more often.

This matters for real work. An early partner reported that Fable 5 “reflects on and validates its own work. For us, that’s what makes highly autonomous operations possible, the extra thinking pays for itself.”

Benchmarks: how Fable 5 compares

Anthropic published a comparison table showing Fable 5 and Mythos 5 against other leading models. The key takeaway is that Fable 5 leads or matches the state of the art on nearly every test dimension.

Anthropic did not publish detailed percentage scores for every benchmark in their announcement. They relied more on customer testimonials and real-world task descriptions. The partner quotes came from teams at Cursor, GitHub, Stripe, Hebbia, IMC, Replit, Lovable, Cognition, and others, all citing their own internal evaluations.

One partner on a vibe-coding benchmark put it plainly: “Apps that took a hundred prompts a year ago, it now one-shots.”

The safeguards: what gets blocked and why

This is the most unusual part of the release. Anthropic built Fable 5 with safeguards that did not exist at this level before.

Fable 5 comes with a set of classifiers, separate AI systems that scan queries for potential misuse. When a classifier detects something risky, the query is not refused. Instead, it gets routed to Claude Opus 4.8, which handles the response.

The classifiers cover three areas:

1. Cybersecurity. Fable 5 can find and exploit software vulnerabilities at a level that could make cyberattacks substantially easier and cheaper. The classifiers block queries related to exploitation and offensive cyber tasks. In internal testing, external red-teaming organizations and a bug bounty program with over 1,000 hours of testing found no universal jailbreaks that could bypass these safeguards.

Anthropic also published an internal evaluation showing Fable 5’s jailbreak resistance against automated red-teaming compared to earlier models:

2. Biology and chemistry. Fable 5 can reason about biological problems at a level that could be misused for bioweapons research. For example, it outperformed specialized protein language models at predicting how genetic modifications would affect viral shell assembly, a task relevant to gene therapy but also to designing dangerous viruses. Until Anthropic narrows these classifiers, Fable 5 falls back to Opus 4.8 on most biology and chemistry queries.

3. Distillation. Anthropic has seen large-scale attempts by actors in authoritarian countries to extract or “distill” Claude’s capabilities to train competing models. Queries flagged as part of such attempts fall back to Opus 4.8.

The safeguards are deliberately conservative. Anthropic admits they are stricter than ideal, and sometimes harmless requests will trigger them. The company said this happens in less than 5 percent of sessions on average, and they plan to reduce false positives after launch.

For the 95 percent of sessions where no classifier triggers, Fable 5 performs the same as Mythos 5.

A new data retention policy also applies to Fable 5, Mythos 5, and future models at a similar capability level. Anthropic will retain business customer traffic data for 30 days, strictly for safety purposes like detecting new jailbreaks and reducing false positives. The data will not be used for training.

Claude Mythos 5: the unrestricted sibling

Claude Mythos 5 is the same underlying model, but with cybersecurity safeguards lifted. It is available only to existing Project Glasswing partners, including the US government and a small group of cybersecurity and critical infrastructure organizations.

Anthropic plans to expand access through a broader trusted access program. It also intends to open a separate trusted access program for biology, giving vetted researchers access to Fable 5 with the biology and chemistry classifiers removed.

In internal testing, Mythos 5 accelerated parts of the drug design process by around ten times. When tested on protein design tasks with no human assistance, it matched or beat skilled human operators. It also produced novel molecular biology hypotheses that Anthropic’s scientists preferred over Opus-class models 80 percent of the time in blinded comparisons.

One Mythos hypothesis, a novel mechanism for an E. coli protein, was independently corroborated by a separate lab working on the same problem.

Mythos 5 also conducted a week of largely autonomous genomics research, assembling single-cell data for millions of cells across 138 animal species and designing a custom machine learning model that outperformed a recent publication in the journal Science, despite being 100 times smaller.

Pricing and availability

Both Fable 5 and Mythos 5 cost $10 per million input tokens and $50 per million output tokens. That is less than half the price of Claude Mythos Preview, but double the cost of Claude Opus 4.8 ($5/$25).

Developers can use the model ID claude-fable-5 via the Claude API. If you want to try it without a direct Anthropic contract, check our OpenRouter free models guide for routing options (availability may lag behind the direct API). On Amazon Bedrock, Fable 5 is available in the US East (N. Virginia) and Europe (Stockholm) regions. The Claude Platform on AWS covers North America, South America, Europe, and Asia Pacific.

For subscription plans, the rollout is staged:

• June 9 to June 22: Fable 5 is included at no extra cost on Pro, Max, Team, and seat-based Enterprise plans.
• June 23 onward: Fable 5 moves to a usage credit model on those plans. If capacity allows, Anthropic will extend the free window.
• Later: Anthropic aims to restore Fable 5 as a standard part of subscription plans once sufficient capacity is available.

API and consumption-based Enterprise plans get full access from day one.

Anthropic expects demand to be very high and difficult to predict. The staged subscription rollout is designed to avoid crashing the service, not to extract more money.

Fable 5 vs Opus 4.8: when to switch

If you use Claude regularly, here is the practical question: when should you pay more for Fable 5 instead of staying on Opus 4.8?

The pricing difference is meaningful. At $10/$50 per million tokens, Fable 5 costs exactly double Opus 4.8. If your workload does not need the extra capability, Opus 4.8 is the better value.

Fable 5 vs the rest of the frontier (GPT-5.5, DeepSeek V4 Pro, Gemini 3.1)

Fable 5 does not exist in a vacuum. The Claude Fable 5 vs GPT-5.5 question is the first one most teams will ask, since those two models sit at the top of the coding and reasoning leaderboards. The frontier AI market has three other heavy hitters, and picking between them comes down to your specific stack and priorities.

Pick Fable 5 if you are already in the Claude ecosystem and hitting quality ceilings on Opus 4.8. The coding and reasoning jump is real, and the price increase over Opus 4.8 is the cost of that ceiling breaking.

Pick GPT-5.5 if your team lives in OpenAI’s world: Codex, ChatGPT, Azure. The agent stack (terminal, browser, spreadsheets) is better integrated than Anthropic’s equivalent. But you will pay more per token and you get a different philosophical approach to safety.

Pick DeepSeek V4 Pro if cost is your primary concern. At roughly one-fifth the price of Fable 5 for outputs, it is the budget frontier option. The 1M token context window is also larger. But you trade off some raw capability on the hardest tasks, especially long and complex ones. Also check your compliance requirements before using DeepSeek for regulated work.

Pick Gemini 3.1 Pro if your files, email, and docs already live in Google Workspace. The 2M token context window is the largest among these models. Multimodal input (video, image, audio) is stronger than Claude’s offering. But for sustained autonomous coding agents, Fable 5 and GPT-5.5 both pull ahead.

The short version: Fable 5 is the new best for autonomous coding agents and hard analytical work if you can afford the token cost. GPT-5.5 is the safer pick for teams already on OpenAI. DeepSeek V4 Pro is the cost leader. Gemini 3.1 Pro is the Google ecosystem play.

Early reactions from developers and enterprises

The announcement landed on June 9, 2026, so broad community reaction is still developing. What we have so far comes from the early access partners Anthropic quoted in their announcement.

Cursor’s CEO Michael Truell said Fable 5 “opened up a class of long-horizon problems that were out of reach for earlier models.”

GitHub CPO Mario Rodriguez described it as “a real step forward” and pointed to the direction it signals: “a future where developers can hand increasingly ambitious work to agents and trust the results across the software lifecycle.”

Replit reported “the strongest results of any Claude model we’ve had the opportunity to test.”

Lovable noted that Fable 5 “understands what builders mean, not just what they type.” Apps that used to take a hundred prompts a year ago now get built in a single shot.

The legal industry also weighed in. One law firm partner said Fable 5 “feels materially different” and that in blind review, its redlines matched or beat their current model every time.

On the analyst side, Fable 5 “is the first to break 90 percent” on one team’s core analytics benchmark, a 10-point jump over Opus 4.8.

The main concern voiced so far is the conservative safeguards sometimes catching harmless requests. Anthropic acknowledges this and says reducing false positives is a priority.

On Hacker News, the initial discussion thread saw some confusion because the announcement page briefly returned a 404 error before going live properly. A few users noted seeing screenshots of the post on Reddit and Twitter before the official page was accessible, suggesting a coordinated launch with a short embargo break.

What this means for the AI model landscape

Fable 5 changes the frontier. Before this release, the most capable models available to anyone were the Opus-class models (Claude Opus 4.8), GPT-5.5, and DeepSeek V4 Pro (see our DeepSeek V4 vs ChatGPT vs Claude comparison). Fable 5 moves the ceiling up by a meaningful margin, especially for long and complex tasks.

For developers building on Claude, the practical effect is clear: if you are currently running Opus 4.8 and hitting quality limits, Fable 5 is the step up. The API integration path is identical. You change the model ID and you get the new capability ceiling.

For teams using GPT-5.5, the Claude Fable 5 vs GPT-5.5 comparison becomes more interesting. Anthropic’s partner reports claim Fable 5 matched GPT-5.5’s frontier physics research output in 36 hours that took GPT-5.5 four days, using a third of the reasoning tokens. That is one data point from one domain, but it suggests the model gap is real.

For the broader ecosystem, the bigger story is the safety framework. Anthropic released a model it previously said was too dangerous to make public. It did so by building a new layer of AI guardrails that inspect every query before the main model responds. Whether this approach works at scale, and whether other labs copy it, will be one of the defining questions of the next few months.

Who should use, watch, or skip

Changelog

• 2026-06-09: First publish. Claude Fable 5 and Claude Mythos 5 release coverage, benchmarks, pricing, safeguards, and early partner reactions.